MIC training: Modern data analysis in R/RStudio

PBS Crash Course CS

A variable refers to a storage location in computer's memory, e.g.

myVariable = 5.3

The symbolic name myVariable refers to a memory location that stores the number 5.3.

A variable can vary!

myVariable = myVariable + 2.8

We changed the value referred by the mnemonic myVariable. Now it stores 8.1.

Data stored under variables can have different types. There are 5 of them in R. Use function typeof() to check.

Taken from R tutorial on TechVidvan.

A data structure is a way of storing and organising data. For example, in order to store 10 integers, we could define 10 variables, which isn't very efficient. Instead, we can store these numbers in a vector.

Taken from R tutorial on TechVidvan.

Control structures change the flow of the code. The changes are based on conditions, e.g. if variable a is greater than a certain value, do this, otherwise, do that.

Taken from R tutorial on TechVidvan.

Stick to a single naming convention throughout the code. A convenient convention is a so-called camel notation, where names of variables, constants, functions are constructed by capitalizing each comound of the name, e.g.:

• calcStatsOfDF - function to calculate stats
• nIteration - prefix n to indicate an integer variable
• fPatientAge - f to indicate a float variable
• sPatientName - s to indicate a string variable
• vFileNames - v for vector
• lColumnNames - l for a list

If a variable is available/valid only within a function, prefix the name with loc, e.g.:

• locNiter
• locVfileNames

The general form of data.table syntax is as follows:

DT[i, j, by]

##   R:                 i                 j        by
## SQL:  where | order by   select | update  group by

The way to read it (out loud) is: take DT, subset/reorder rows using i, then calculate j, grouped by by.

Select specific records:

dtBabies[gender == 'M']

      name gender year2011 year2012 year2013 year2014 year2015
1: Jackson      M    74.69    84.99    91.73    95.32   107.12
2:    Liam      M    88.24       NA   101.83   108.23   119.01

Select specific columns:

dtBabies[, .(name, gender, year2015)]

      name gender year2015
1: Jackson      M   107.12
2:    Emma      F    93.73
3:    Liam      M   119.01
4:     Ava      F   105.65

Select specific records:

dtBabies[get(lCol$group[[2]]) == 'M']

      name gender year2011 year2012 year2013 year2014 year2015
1: Jackson      M    74.69    84.99    91.73    95.32   107.12
2:    Liam      M    88.24       NA   101.83   108.23   119.01

Select specific columns:

myColumns = c(lCol$group[[1]], lCol$timeLast)
dtBabies[, ..myColumns]

      name year2015
1: Jackson   107.12
2:    Emma    93.73
3:    Liam   119.01
4:     Ava   105.65

Our data table is in the wide format. To convert it to long format, use the function melt.

Provide the names of identification (id.vars) and measure variables (measure.vars). If none are provided, melt guesses them automatically, which may result in a wrong conversion. Both variables can be given as strings with column names, or as column numbers.

The original data frame contains missing values; na.rm=T omits them in the long-format table.

dtBabiesLong = data.table::melt(dtBabies, 
                                id.vars = c('name', 'gender'), measure.vars = 3:7,
                                variable.name = 'year', value.name = 'weight',
                                na.rm = T)
head(dtBabiesLong, n = 5L)

      name gender     year weight
1: Jackson      M year2011  74.69
2:    Liam      M year2011  88.24
3:     Ava      F year2011  81.77
4: Jackson      M year2012  84.99
5:     Ava      F year2012  96.45

The function dcast converts from long to wide format. The function has a so called formula interface that specifies a combination of variables that uniquely identify a row.

Note that because some combinations of name + gender + year do not exist, the dcast function will introduce NAs.

dtBabiesWide = data.table::dcast(dtBabiesLong, 
                                 name + gender ~ year, 
                                 value.var = 'weight')

dtBabiesWide

      name gender year2011 year2012 year2013 year2014 year2015
1:     Ava      F    81.77    96.45       NA       NA   105.65
2:    Emma      F       NA       NA    75.74    82.49    93.73
3: Jackson      M    74.69    84.99    91.73    95.32   107.12
4:    Liam      M    88.24       NA   101.83   108.23   119.01

One way to use the formula interface with string variables is to create the formula string explicitly by concatenating individual string pieces:

as.formula(
  paste0(
    lCol$group[[1]], "+", 
    lCol$group[[2]], "~", 
    lCol$time))

name + gender ~ year

p1 = ggplot2::ggplot(dtBabiesLong, 
                     aes_string(x = lCol$time, 
                                y = lCol$meas, 
                                group = lCol$group[1])) +
  geom_line() +
  geom_point() +
  facet_wrap(sFormula) +
  stat_summary(fun.y = mean, 
               aes(group=1), 
               geom = "line", 
               colour = 'red')

The group=1 overrides the by-name grouping so that we get a mean across all names for each gender, rather than the mean of each individual name within each gender (which would be the same as individual observations).

Making an interactive plot from a ggplot object is extremely easy. Just use ggplotly function from the amazing plotly package. The interactive plot will remain in the html document knitted from the R notebook.

require(plotly)
plotly::ggplotly(p1)

Let's write a function to calculate statistics of a data frame. In the simplest case the function will calculate the mean of a single column of a data frame.

We will expand the function to calculate the mean by a group, and to calculate robust statistics, i.e. median instead of the mean.

The function will need the following input parameters:

• the name of the data frame to use for calculations
• the name of the variable to summarise
• the name of the column with grouping (optional)
• a True or False parameter for robust stats (optional)

calcStats = function(inDt, 
                     inMeasVar) {

  outDt = inDt[, 
               .(meanMeas = mean(get(inMeasVar)))]

  return(outDt)
}

Since column names will be provided to our function as string parameters, we cannot hard-code them inside of the function. Therefore, we use function get to use the string stored in variable inMeasVar as the column name.

Let's apply the function to our dtBabiesLong table:

calcStats(inDt = dtBabiesLong, 
          inMeasVar = "weight")

   meanMeas
1: 93.79933

Extension 1: allow for calculating the mean by group via an optional parameter:

calcStats = function(inDt, 
                     inMeasVar, 
                     inGroupName = NULL) {

  outDt = inDt[, 
               .(meanMeas = mean(get(inMeasVar))), 
               by = inGroupName]

  return(outDt)
}

calcStats(inDt = dtBabiesLong, 
          inMeasVar = "weight",
          inGroupName = "gender")

   gender meanMeas
1:      M 96.79556
2:      F 89.30500

If the parameter inGroupName is not provided, the mean will be calculated for the entire table without grouping.

Extension 2: add a parameter inRobust to select between calculating the mean and the median.

calcStats = function(inDt, 
                     inMeasVar, 
                     inGroupName = NULL, 
                     inRobust = F) {

  if (inRobust) {
    outDt = inDt[, 
                 .(medianMeas = median(get(inMeasVar))), 
                 by = inGroupName]
  } else {
    outDt = inDt[, 
                 .(meanMeas = mean(get(inMeasVar))), 
                 by = inGroupName]
  }

  return(outDt)
}

calcStats(inDt = dtBabiesLong, 
          inMeasVar = "weight",
          inGroupName = "gender",
          inRobust = T)

   gender medianMeas
1:      M      95.32
2:      F      88.11

Possible extensions could involve error handling, e.g.:

• Check whether column names provided via inMeasVar and inGroupName exist in the data table.
• Check whether the data table is not empty.

Once inside the function, click Menu > Code > Insert Roxygen Skeleton (Shift-Option-Command R). A special type of comment will be added above the function. You can add your text next to parameters.

#' Calculates stats of a data frame
#'
#' @param inDt Input data table in the long format
#' @param inMeasVar Name of the measurement column
#' @param inGroupName Name of the grouping column (default NULL)
#' @param inRobust If true, the function calculates median instead of the mean (default False)
#'
#' @return Data table with summary stats
#' @export
#' @import data.table
#'
#' @examples
#' # example usage 

calcStats = function(inDt, inMeasVar, inGroupName = NULL, inRobust = F) {

  if (inRobust) {
    outDt = inDt[, .(medianMeas = median(get(inMeasVar))), by = inGroupName]
  } else {
    outDt = inDt[, .(meanMeas = mean(get(inMeasVar))), by = inGroupName]
  }

  return(outDt)
}

calcStats = function(inDt, inMeasVar, inGroupName = NULL, inRobust = F) {
  require(data.table)

  if (is.null(inDt))
    stop("input data.table is NULL")

  if(!is.data.table(inDt))
    stop("input is not a data.table")

  if(nrow(inDt) == 0)
    stop("input data.table has 0 rows")

  if(!(inMeasVar %in% names(inDt)))
    stop("column name does not exist in the data.table")

  if(!is.logical(inRobust))
    stop("argument inRobust must be logical, TRUE or FALSE")

  if (inRobust) {
    outDt = inDt[, .(medianMeas = median(get(inMeasVar))), by = inGroupName]
  } else {
    outDt = inDt[, .(meanMeas = mean(get(inMeasVar))), by = inGroupName]
  }

  return(outDt)
}

require(testthat)

# calculated result
resCalc = data.table(meas = 1:9)
resTrue = data.table(meanMeas = 5.0)

# Test should pass; NO message will be produced
testthat::expect_equal(calcStats(inDt = resCalc,
                                 inMeasVar = "meas"), 
                       resTrue)

# Test should not pass; error message will appear
resTrue = data.table(meanMeas = 4.0)

testthat::expect_equal(calcStats(inDt = resCalc,
                                 inMeasVar = "meas"), 
                       resTrue)

Error: calcStats(inDt = resCalc, inMeasVar = "meas") not equal to `resTrue`.
Column 'meanMeas': Mean relative difference: 0.2

Create a test folder and create individual files with tests. The name of these files should start with test_ prefix, e.g. test_calcStats.R.

require(testthat)

# Source the file with the function to test; adjust the path if required.
source("../calcStats.R")

testthat::test_that("tests", {

  expect_error(calcStats(NULL, "x"))
  expect_equal(...)
})

Run tests on the entire tests folder with testthat::test_dir("examples/tests") or on individual test files with testthat::test_file("examples/tests/test_calcStats.R").

The output should like similar to this:

> testthat::test_file("examples/tests/test_calcStats.R")
✓ |  OK F W S | Context
✓ |   8       | test_calcStats

══ Results ══
OK:       8
Failed:   0
Warnings: 0
Skipped:  0

Profiling with RStudio.

The profiler is a tool for helping you to understand how R spends its time. It provides a interactive graphical interface for visualizing data from:

• Rprof, R’s built-in tool for collecting profiling data and,
• profvis, a tool for visualizing profiles from R.

To make slow code faster, we need accurate information about what is making our code slow.

Source code in ./examples/ex_profiling_time01.R

require(profvis)

# Define data size
nRows = 4e5
nCols = 150

# Create data
# Random Gaussian numbers with mean = 5 , sd = 1
myData = matrix(rnorm(nRows * nCols, 
                      mean = 5),
                ncol = nCols)

# Start profiling
profvis({
  # Get column means: 2 methods
  # 1. Apply the mean function to every column
  locColMeans = apply(myData, 2, mean)

  # 2. use colMeans function on the entire matrix
  locColMeans = colMeans(myData)

  # Loop over each column.
  # subtract mean from each column.
  for (ii in seq_along(locColMeans)) {
    myData[, ii] <-
      myData[, ii] - locColMeans[ii]
  }
})

Use system.time() function to measure the elapsed time.

nRows = 4e4
nCols = 100

myData = matrix(rnorm(nRows * nCols, 
                      mean = 5),
                ncol = nCols)

locColMeans = colMeans(myData)

system.time(
  for (ii in seq_along(locColMeans)) {
    myData[, ii] <-
      myData[, ii] - locColMeans[ii]
  }
)

   user  system elapsed 
  0.080   0.041   0.154 

Use microbenchmark package to accurately measure and compare the execution time of R expressions.

Source code in ./examples/ex_profiling_time02.R

# Benchmark 3 approaches to center columns:
# 1. using base R, loop over columns and subtract precomputed means,
# 2. using data.table with lapply,
# 3. using data.table, loop over columns and use set() function
mbm = microbenchmark(
  base_r = for (ii in seq_along(myColMeans)) {
    data1[, ii] <-
      data1[, ii] - myColMeans[ii]},

  dt_lapply = data2[, 
                    (myColNames) := lapply(.SD, 
                                           function(x) x - mean(x)), 
                    .SDcols = myColNames],

  dt_for_set = for (ii in myColNames) 
    set(data3, 
        j = ii, 
        value = data3[[ii]] - mean(data3[[ii]])),

  times = 100L)

# Display benchmark table
mbm

Unit: milliseconds
expr      min       lq     mean  median       uq       max neval cld
base_r 672.9243 716.4792 878.6226 851.141 921.6990 1529.4567   100   b
dt_lapply 347.4257 367.2748 469.3408 388.986 513.9202  944.1644   100  a 
dt_for_set 336.9143 367.0250 445.6910 376.871 499.6137  943.2988   100  a 

# Plot results of microbenchmark
ggplot(mbm) +
  geom_violin(aes(x = expr, 
                  y = time)) +
  coord_flip()

Debugging with RStudio.

Debugging is designed to help you find bugs by figuring out where the code is not behaving in the way that you expect. To do this, you need to:

1. Begin running the code,
2. Stop the code at the point where you suspect the problem is arising,
3. Look at and/or walk through the code, step-by-step at that point.

As an alternative to for loops, try using functions from the apply family, e.g. lapply, sapply, etc. The advantage is that it forces you to encapsulate your code into functions, which makes the code reusable and modular.

For example, we have several CSV files in examples/ex_apply_data folder that we would like to read and combine into a single data.table:

require(data.table)

# List all CSV files in the folder; store them in a character vector
vFiles = list.files(path = "examples/ex_apply_data/.", 
                    pattern = "csv", 
                    full.names = T)

# Pre-allocate a list which will hold data tables from individual files
lMyExp = list()

# Loop over file names and read them
for (ii in seq_along(vFiles)) {
  lMyExp[[ii]] = fread(vFiles[[ii]])
}

# Combine the list into a data.table
dtMyExp1 = rbindlist(lMyExp)

A simpler way is to use the lapply function that loops over all elements of the vFiles vector and applies a function data.table::fread. As previously, the result is stored in a list. Example code.

lMyExp = lapply(vFiles, fread)
dtMyExp2 = rbindlist(lMyExp)

all.equal(dtMyExp1, dtMyExp2)

[1] TRUE

The main reason for using the foreach package is that it supports parallel execution; it can execute repeated operations on multiple processors/cores on your computer, or on multiple nodes of a cluster.

R notebook with example calculations.

Template code where the result of each iteration is stored in a myResult list:

require(foreach)
require(doParallel)

numCores = 4
doParallel::registerDoParallel(numCores)

system.time({
  myResult = foreach::foreach(ii=1:10,
                              .packages='...') %dopar%

    myFunction(ii)
  ...
})
doParallel::stopImplicitCluster()

Here, the results of each iteration are combined into a vector:

system.time({
  myResult = foreach::foreach(ii=1:10,
                              .combine = 'c',
                              .packages='...') %dopar%

    myFunction(ii)
  ...
})

Parse command-line arguments using the optparse package (code taken from R-bloggers)

Here, we create two command-line parameters -r and -d. Both have their respective options in the long format, --rootdir and --debug.

The default value for -r is NULL which, together with a subsequent if-stop statement, makes it a required parameter. The -d has a default value set to FALSE.

#!/usr/bin/env Rscript

require(optparse)

option_list = list(
  optparse::make_option(c("-r", "--rootdir"), type="character", default=NULL, 
                        help="full path to root directory", metavar="character"),
  optparse::make_option(c("-d", "--debug"), action="store_true", default = FALSE, dest = 'debug', 
                        help="Print extra output")
); 

opt_parser = optparse::OptionParser(option_list=option_list)
opt = optparse::parse_args(opt_parser)

The list opt contains all the arguments sorted by order of appearance in option_list and which can be called by their names as declared in the object: opt$rootdir, opt$debug, etc.

Manage NULL arguments as follows:

if (is.null(opt$rootdir)){
  optparse::print_help(opt_parser)
  stop("Please supply a full path to root directory with data to analyse.\n", call.=FALSE)
}

l.par$dir.root = opt$rootdir
cat(sprintf('Working in:\n%s\n\n', l.par$dir.root))

if (opt$debug)
  cat('\nEntering debug mode. More output...\n\n')

The source code for this example is in examples/ex_opparse.R.

Thanks to the first line (#!/usr/bin/env Rscript), a so-called shebang, you can run the script by simply executing ./ex_optparse.R in the command line. Otherwise, you can type Rscript ex_optparse.R.

This is the output, if no parameters given:

Loading required package: optparse
Usage: ./ex_optparse.R [options]


Options:
    -r CHARACTER, --rootdir=CHARACTER
        full path to root directory

    -d, --debug
        Print extra output

    -h, --help
        Show this help message and exit


Error: Please supply a full path to root directory with data to analyse.
Execution halted